GatorLM

GatorLM is a custom GPT-style language model developed at the University of Florida.

Model Details

  • Architecture: Custom GPT with RoPE, GQA, RMSNorm, and SwiGLU MLP
  • Parameters: ~2B
  • Context length: 2048 tokens
  • Tokenizer: GPT-2 (tiktoken)
  • Dtype: bfloat16

Usage

This model uses a custom inference handler. Send requests in the following format:

from huggingface_hub import InferenceClient

client = InferenceClient(model="Krill11/GatorLM1")
response = client.post(json={"turns": [], "message": "Hello, who are you?"})

Input format

{
  "turns": [["previous user message", "previous assistant reply"]],
  "message": "current user message"
}
  • turns: list of completed [user, assistant] exchange pairs (empty list for a fresh conversation)
  • message: the new user message

Output format

{
  "reply": "GatorLM's response"
}

Training

Fine-tuned via supervised fine-tuning (SFT) on conversational data using the Muon optimizer.

Downloads last month

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support